AgentsSeptember 13, 2026via The Decoder
Iris-mini and Iris-pro are the strongest open-weight search agents in their class
Why it matters
New open-weight search agents (Iris-mini, Iris-pro) deliver frontier-class performance in their size classes and show unexpected generalization to tool use and office work — a significant capability checkpoint for practitioners evaluating open alternatives to proprietary agent platforms.
Key signals
- AllSpark released Iris-mini and Iris-pro open-source search agents
- Built on Qwen models
- Lead benchmarks among open-weight models in their respective size classes
- Training data and models improve performance on untrained tasks (general tool use, office work)
- Demonstrates transfer learning across agent capabilities
- Iris-mini and Iris-pro released as open-source search agents
- Show improved performance on untrained tasks (general tool use, office work)
- Training data and model weights publicly available
The hook
AllSpark's Iris agents top open-weight benchmarks — and transfer surprisingly well to tasks they never saw.
The AllSpark team has released Iris-mini and Iris-pro, two open-source search agents built on Qwen models that lead benchmarks among open-weight models in their size classes. According to the paper, the training data and models also improved performance on tasks they were never trained for, includin…